Yu Wang

Marco Manfredi, Yu Wang

Robustness to small image translations is a highly desirable property for object detectors. However, recent works have shown that CNN-based classifiers are not shift invariant. It is unclear to what extent this could impact object detection, mainly because of the architectural differences between the two and the dimensionality of the prediction space of modern detectors. To assess shift equivariance of object detection models end-to-end, in this paper we propose an evaluation metric, built upon a greedy search of the lower and upper bounds of the mean average precision on a shifted image set. Our new metric shows that modern object detection architectures, no matter if one-stage or two-stage, anchor-based or anchor-free, are sensitive to even one pixel shift to the input images. Furthermore, we investigate several possible solutions to this problem, both taken from the literature and newly proposed, quantifying the effectiveness of each one with the suggested metric. Our results indicate that none of these methods can provide full shift equivariance. Measuring and analyzing the extent of shift variance of different models and the contributions of possible factors, is a first step towards being able to devise methods that mitigate or even leverage such variabilities.

Kumpati S. Narendra, Snehasis Mukhopadyhay, Yu Wang

This is the first of a series of papers that the authors propose to write on the subject of improving the speed of response of learning systems using multiple models. During the past two decades, the first author has worked on numerous methods for improving the stability, robustness, and performance of adaptive systems using multiple models and the other authors have collaborated with him on some of them. Independently, they have also worked on several learning methods, and have considerable experience with their advantages and limitations. In particular, they are well aware that it is common knowledge that machine learning is in general very slow. Numerous attempts have been made by researchers to improve the speed of convergence of algorithms in different contexts. In view of the success of multiple model based methods in improving the speed of convergence in adaptive systems, the authors believe that the same approach will also prove fruitful in the domain of learning. In this paper, a first attempt is made to use multiple models for improving the speed of response of the simplest learning schemes that have been studied. i.e. Learning Automata.